The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill instructs the user or agent to download, compile, and install a binary from an external, third-party GitHub repository (https://github.com/f/mcptools.git) which is not controlled by a recognized trusted organization.
Evidence: The 'Prerequisites' section contains commands to git clone the repository and run go build to create a local binary.
[EXTERNAL_DOWNLOADS]: The skill frequently uses npx -y and docker run to download and execute various Model Context Protocol (MCP) servers from public registries (NPM and GHCR).
Evidence: Examples include npx -y @modelcontextprotocol/server-filesystem and docker run -i --rm ghcr.io/github/github-mcp-server.
[COMMAND_EXECUTION]: The core functionality of the skill is centered around executing shell commands to discover and invoke tools provided by external servers.
[CREDENTIALS_UNSAFE]: The skill provides instructions and examples for handling sensitive authentication data (passwords, tokens, API keys) via command-line flags and environment variables, which can expose secrets in shell history, process listings, or logs.
Evidence: Examples include --auth-user "username:password", --auth-header "Bearer your-token-here", and setting BRAVE_API_KEY.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and act upon data (tools, resources, and prompts) provided by external MCP servers, creating an attack surface where a malicious server could influence the agent's behavior.
Ingestion points: Content returned from mcp tools, mcp resources, and mcp prompts calls.
Boundary markers: None identified; server-provided content is integrated into the workflow.
Capability inventory: The skill allows the agent to execute tools, read resources, and apply prompt templates provided by the server.
Sanitization: No mention of sanitization or validation of the schemas or content returned by external servers.

mcp-cli